AI tools for small business: what to use, by category

Start with categories, not product names

Lists of "the 50 best AI tools" are mostly noise for a 20-person firm. What helps is knowing which category of tool solves which kind of problem, then choosing within that category based on the systems you already run. Four categories cover almost everything a small service business needs:

Most firms need something from the first three categories and only occasionally the fourth. Working through them in that order keeps costs down and avoids buying a specialist tool for a job a general assistant handles well.

• General AI assistants for drafting, summarizing and analysis.
• AI features built into your existing business software.
• Automation platforms that move data between systems and add AI steps.
• Specialist AI products for one job, such as call answering or document extraction.

General AI assistants

These are the chat-style tools most people know: ChatGPT, Claude, Microsoft 365 Copilot and Google Gemini. They handle drafting, rewriting, summarizing, brainstorming and working through documents. For business use, buy a business or team plan under a company account so you control access and get business data terms.

Published prices change, so check before you buy. As of November 2026, Claude Team listed $20 per seat per month on annual billing, and Microsoft 365 Copilot Business was listed from $18 per user per month on annual terms, on top of a Microsoft 365 plan.

Choosing between them is often simpler than it looks. If your firm runs on Microsoft 365, Copilot has the advantage of working inside Outlook, Word and Teams with your existing permissions. If you run on Google Workspace, Gemini fits similarly. Otherwise, trial two assistants on the same five real tasks and let the people who will use them decide.

Whichever you choose, standardize on one for the company. Supporting several general assistants multiplies training, policy and security work for little gain.

AI inside the software you already own

Vertical software vendors have been adding AI features quickly. Practice management, field service, CRM, accounting and property management platforms increasingly offer AI summaries, drafting, call handling or data extraction as included features or paid add-ons.

These are worth checking first for two reasons. The data is already there, so there is no new integration to build or maintain. And the vendor's existing data terms and security controls usually apply. The tradeoff is that built-in features can be shallow, and add-on pricing varies widely. Ask your vendor for a demo on your own data and the current add-on price, and ask what happens to your data when the feature runs.

Turn features on deliberately rather than all at once. A single feature, such as AI call summaries in your phone system or draft replies in your help desk, is easier to evaluate and train on than a whole bundle.

Automation platforms

Automation platforms connect your tools so information moves without retyping. Zapier is the most widely known and the easiest to start with. Make suits more complex, branching workflows. n8n appeals to teams that want self-hosting or more technical control. All three can add AI steps, such as classifying an email or drafting a reply, inside a workflow.

As of November 2026, Zapier's pricing page listed paid plans starting at $19.99 per month, with team plans higher and cost rising with task volume. The bigger cost is usually setup and upkeep: someone has to design the workflow, handle errors and fix it when a form or field changes.

If no one in the firm is comfortable maintaining automations, choose the simplest platform and the fewest steps, or budget for outside help with upkeep. An automation nobody understands is a liability.

Specialist AI products

Specialist tools focus on one workflow and often do it better than a general assistant. Common categories in service businesses include AI call answering and intake, meeting transcription and notes, document extraction for invoices and forms, proposal and quote builders, and website chat assistants.

They are worth it when the workflow is frequent and valuable, and the product integrates with your system of record. Before buying, ask for a trial on your real calls, documents or enquiries and measure accuracy yourself. Check the data terms, especially for recordings and client documents. And confirm you can export your data if you leave.

Be cautious with long contracts. Specialist AI products are changing quickly, and a tool that leads its category today may be overtaken next year. Month-to-month or short terms are worth a small premium.

How to choose without a long evaluation

A light process works for most firms. Write the job to be done in one sentence. List the systems the tool must work with. Shortlist two or three options, including any feature in software you already own. Trial each on the same five to ten real examples, with the people who will use it. Check data terms against your AI policy. Then decide, and set a 90-day review.

Write the result on one page: what you chose, why, the cost and the date you will review it. That page saves a repeat evaluation when someone suggests a new tool six months later.

• Does it work with our system of record, such as our CRM, practice management or accounting tool?
• Is there a business plan with admin controls and clear data terms?
• How accurate was it on our own examples, and how long did corrections take?
• What does it cost at our size in year one and year two, including setup?
• Can we export our data and switch it off cleanly?

Avoiding tool sprawl

The most common outcome of AI enthusiasm in small firms is a pile of overlapping subscriptions: three meeting note-takers, two writing assistants and a chatbot nobody maintains. Each is cheap; together they add cost, confusion and data risk.

Keep an approved tool list, review subscriptions every six months, and retire anything nobody uses. InTheory often starts with exactly this kind of software stack review, because cutting unused tools can fund the ones that actually help.

When someone asks for a new AI tool, ask first whether an approved one already covers the need. Often the answer is yes with a better prompt or a setting change. If not, run the same light evaluation and add the winner to the list, retiring anything it replaces.